Computationally optimized versions for mean(), std(), and var()

The Mathworks-supplied functions for mean(), var() and std() calculation (especially the latter two) are severely suboptimal in many cases where the calculation is performed on a matrix without NaN values. While the Mathworks-supplied versions are versatile, this makes them slow. I therefore made condensed versions that accept only two inputs: an ND input matrix (first argument) and an integer specifying which dimension to calculate the metric over (second argument).

The resulting speed-up is around 50% of original computation time for xvar() and xstd(), and 95% for xmean(), but mileage may vary. The results of xvar(), xmean() and xstd() are the same as var(), mean(), and std() up to machine precision.
By Dr. Jorrit S. Montijn, 29-11-16 (dd-mm-yy; Universite de Geneve)
=======================
The included .pdfs show the performance of the original versions versus the updated versions, according to the Matlab Profiler on my personal machine.
Disclaimer: I've tested the performance increases only on my personal PC, so you may experience better or worse improvements depending on your setup.